iMOD User Manual version 4.4 (html)

10.6Data Set 5: Solver configuration

Data Set 5
(optional PCG)
(optional PKS)



Maximal number of “outer” iteration loops. The iterative procedure used in MODFLOW for solving nonlinear problems is commonly referred to as Picard iteration. It splits the solving process into an outer iteration loop, the equation that needs to be solved in the “inner” is (re)formulated.
If OUTER < 0, the Parallel Krylov Solver (PKS) is activated instead of PCG, and for this solver OUTER = -OUTER is taken as the number of outer iterations. For PKS, the supported preconditioning is Incomplete LU-factorization only, which is automatically being set. Hence, you are not allowed to set NPCOND when PKS is activated.


Maximal number of “inner” iteration loops. Within each inner iteration loop, the equation that was formulated in the outer iteration loop is (partly) solved. In common, it is more expensive to have much inner iteration though each iteration loop represents a temporary formulation of the equation. True linear systems can speed up drastically by increasing the number of inner iterations and RELAX \(=1\), however, most groundwater models are a mixture of linear and non-linear elements and therefore a fair trade-off between robustness and speed seems to be NITER \(=20\).



Closure criterion for the hydraulic head (state variable). Commonly it is practice to choose HCLOSE to be 2 orders of magnitude smaller than the desired accuracy to be obtained.



Closure criterion for the mass balance. This criterion depends on the grid size, since large grid cells produce larger errors in mass balances than smaller ones does.


Relaxation factor that quantifies the amount of confidence for each solution obtained after an inner iteration loop, default value is 0.98. It influences the robustness and efficiency of convergence. For purely linear systems it can/must be 1.0, though non-linear system prefers lower values (e.g. 0.50-0.97). It is difficult to know the optimal value for RELAX beforehand. Use the adaptive damping (IDAMPING) instead.

(optional, PCG only)

Pre-conditioning method. If the Preconditioning Method is set to Cholesky, the Relaxation parameter can be set. Although the default is 1, in some cases a value of 0.97-0.99 may reduce the number of iterations required for convergence.


Modified Incomplete Cholesky (for use on scalar computers)


Polynomial matrix conditioning method (for use on vector computers or to conserve computer memory)


(optional, PCG only)

Maximal overall acceptable error for the water balance in percentage (the default value = 0.0% and the solver stops whenever the criterion of QCLOSE cannot be met). Whenever the external-iteration does not converge (due to numerical instability), the simulation will continue whenever the overall error in de the water balance is less than the given MAXWBALERROR criterion.

(optional, PKS only)

Subdomain partition option. There are two methods supported: PARTOPT \(=0\) (default) for uniform partitioning in lateral x and y-direction; PARTOPT \(=1\) that enables the Recursive Coordinate Bisection partitioning method computes the subdomain dimensions according to a load pointer IDF grid. Note that each subdomain always includes all model layers.


Uniform subdomain partitioning (default)


Recursive Coordinate Bisection (RCB) subdomain partitioning

(optional, PKS only)

Flag for merging parallel subdomain IDF output files. The default is IDFMERGE \(=0\), corresponding to no merging. In this case, each subdomain writes its output IDF files like "<IDFNAME>_p<MPIRANK>.idf", where "<IDFNAME>" is the iMOD output variable name, e.g. "head_steady-state_l1" and "<MPIRANK>" the three-digit subdomain MPI rank identifier. Note that enabling this option could slow down overall parallel computations.


No merging of subdomain IDF output files (default)


Merging of subdomain IDF output files